---
title: "Agent优化器会复合吗？基于Terminal-Bench 2.0的持续学习评估"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.14004v1"
permalink: "https://feed.astralor.com/items/itm_7fb45560e7cac385c7edd981"
published: "2026-07-15T16:36:04.000Z"
score: 78
channel: "ai"
tags: ["Agent优化","GEPA","Meta Harness","RELAI-VCL","Terminal-Bench 2.0","持续学习"]
---

# Agent优化器会复合吗？基于Terminal-Bench 2.0的持续学习评估

## AI 点评

> 这篇论文直击了当前Agent优化评估的最大盲区：一次性测试无法反映真实世界的持续变化。如果你正在构建或部署长期运行的Agent系统，这篇文章值得仔细阅读——RELAI-VCL的回归控制思路可能会成为后续设计的标准组件。对于AI研究者而言，论文的实验对比和发现问题的方法论也很有启发。

## 速览

这篇论文质疑了当前Agent优化方法仅在静态基准上一次性评估的做法，提出了持续学习场景下的两阶段评估框架。在Terminal-Bench 2.0的困难任务上，作者比较了GEPA、Meta Harness和RELAI-VCL三种优化方法。结果显示，只有RELAI-VCL通过内置回归控制实现了优化收益的复合，在新任务上既能正向迁移又能持续提升，终身平均通过率达到76.4%，远超基线和其他方法。

## 关键点

- 现有Agent优化方法多在静态基准上一次性测试，忽略了实际部署中任务持续出现的场景。
- 论文提出两阶段持续学习评估，检验优化收益是否在新任务上复合。
- 实验发现，只有引入回归控制的RELAI-VCL方法能同时实现正向迁移和持续提升。
- RELAI-VCL的终身平均通过率达76.4%，而GEPA和Meta Harness均低于基线或无法进一步优化。

## 全文

Most reported gains from agent-optimization methods are one-shot: an agent is optimized against a fixed benchmark and the resulting improvement is reported as if it were a stable property of the method. This does not test the setting that matters for deployed agents, where optimization is applied recursively as new failures and new tasks appear over time. The central question this raises is whether optimizer-driven gains compound: after an agent has been optimized once, can it be optimized again on newly arrived tasks without eroding the gains the first round produced? We study this question with a two-phase continual-learning evaluation built from hard tasks in Terminal-Bench 2.0, comparing three approaches to agent-harness optimization (GEPA, Meta Harness, and RELAI's Verifiable Continual Learning, RELAI-VCL) under identical optimization budgets. All three methods improve over the baseline agent in the conventional, static, single-phase setting. However, once new tasks are introduced, the methods diverge sharply: GEPA's optimized agent transfers below the unoptimized baseline, Meta Harness transfers well but fails to improve further once given a second optimization budget, and RELAI-VCL is the only method that both transfers positively to unseen tasks and continues improving after those tasks are folded into the optimization objective, reaching the highest pass rate at every evaluated stage and the highest lifelong average pass rate overall (76.4% vs. 66.0% for GEPA, 64.6% for Meta Harness, and 58.7% for the baseline). Our key observation was that optimization gains compounded only when regression control was built into the optimization loop, providing an inductive bias against shortcut solutions that fail to generalize.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_7fb45560e7cac385c7edd981